speckit-tasks

name	speckit-tasks
description	Break down implementation plans into actionable task lists.
compatibility	Requires spec-kit project structure with .specify/ directory
metadata	{"author":"github-spec-kit","source":"preset:fsharp-opinionated"}
user-invocable	true
disable-model-invocation	false

Speckit Tasks Skill

/speckit.tasks

Generate the feature's task breakdown from its spec and plan. This preset requires TWO files, both in specs/[FEATURE_ID]/:

1. tasks.md — the human checklist. Status legend:

   • [ ] pending / [X] done (real evidence) / [S] synthetic-only / [F] failed / [-] skipped.
   • Never write [S*] yourself. That marker is computed by the evidence audit from the DAG; writing it by hand will confuse the script.

2. tasks.deps.yml — the dependency topology. Sibling to tasks.md. Every Tnnn id in tasks.md MUST appear as a key here, even if its deps list is empty.

Use the preset's tasks-template.md and tasks.deps-template.yml as starting points; replace the example task bodies with real work items derived from the spec and plan.

Discipline

• Lockstep emission. You MUST write both files in the same turn. Never emit tasks.md without tasks.deps.yml, and vice versa. If you only have partial information, write placeholder deps (Tnnn: []) and note which ones need review in your summary.
• Phase-checkpoint edges are implicit. The graph compute script auto-injects an edge from every task in Phase N+1 to the last foundation task of Phase N. You do NOT repeat those edges in the yml — write only non-phase cross-edges (e.g., T015: [T011, T013] says US1 impl depends on US1 tests and US1 fixtures).
• Story grouping. Tasks belong to a phase (Phase 1..N) and optionally a user story ([US1], [US2], ...). Keep phases sequential; stories within a phase may run in parallel.
• Tier annotation. Mark each task [T1] or [T2] if the phase classification differs from the spec's overall tier. Omit when it matches.
• Parallel-safe marker. [P] means "no deps inside this phase" — the script can verify it; you SHOULD emit it as a hint.
• Synthetic-evidence inventory. Include the empty Synthetic-Evidence Inventory table from the template. It will grow as /speckit.implement adds [S] tasks.

Validation

Immediately after writing both files, run:

.specify/extensions/evidence/scripts/bash/run-audit.sh --graph-only

(or invoke /speckit.graph.compute if the extension is installed).

This validates:

• Every Tnnn in tasks.md has a matching key in tasks.deps.yml.
• Every dep reference resolves to a known Tnnn.
• The graph is acyclic.

Report any failures to the user immediately; refuse to declare the tasks phase complete until the DAG is clean.

If the evidence extension is not installed

Fall back to emitting both files without running the validator. Warn the user: "The evidence extension is not installed, so the DAG cannot be validated. Run specify extension add evidence to enable speckit.graph.compute and speckit.evidence.audit."